Changelog

The review desk cleared six weeks of proposals. 57 new sources entered the reading library and 9 new claims joined the receipts. Guardrails gains a section on the limits of pre-authored policies, along with evidence for deterministic checks at the point of execution and for the gap between a model recognizing forged authority and acting on that recognition. The cost reality now shows per-unit price and total spend moving independently. The governance gap adds the distance between confidence in passing an audit and actual governance maturity. Tool orchestration extends error recovery to degraded tool payloads as well as failed calls. Two passages describing specific incidents were removed from Guardrails and Prompt injection, so the prose stays about patterns rather than events. Every addition passed the human gate.

The pipeline delivered 101 approved changes since the last entry. 104 new sources entered the reading library. 14 new claims joined the receipts. Concepts touched: The cost reality, The governance gap, Guardrails, Multi-agent topologies, Tool orchestration, Verification loops, ISO/IEC 42001, MCP (Model Context Protocol). Every addition passed the human gate.

The pipeline delivered 1 approved change since the last entry. 1 new source entered the reading library. Every addition passed the human gate.

The pipeline delivered 2 approved changes since the last entry. 2 new sources entered the reading library. 1 new claim joined the receipts. Concepts touched: Verification loops. Every addition passed the human gate.

The pipeline delivered 3 approved changes since the last entry. 3 new sources entered the reading library. Every addition passed the human gate.

The pipeline delivered 4 approved changes since the last entry. 3 new sources entered the reading library. Every addition passed the human gate.

The pipeline delivered 2 approved changes since the last entry. 1 new source entered the reading library. Every addition passed the human gate.

The pipeline delivered 1 approved change since the last entry. 1 new source entered the reading library. Every addition passed the human gate.

The pipeline delivered 1 approved change since the last entry. 2 new sources entered the reading library. 2 new claims joined the receipts. Concepts touched: The regulatory context. Every addition passed the human gate.

The pipeline delivered 9 approved changes since the last entry. 9 new sources entered the reading library. 3 new claims joined the receipts. Concepts touched: Measuring what matters, The scaling gap. Every addition passed the human gate.

The pipeline delivered 19 approved changes since the last entry. 19 new sources entered the reading library. 3 new claims joined the receipts. Concepts touched: Measuring what matters, The scaling gap. Every addition passed the human gate.

The living pipeline completed its first full cycle: everything the field published since the paper's snapshot, read, placed on the map and reviewed by the editor. The reading library grew by 49 sources across all six sides. Four new verified claims entered the receipts, including fresh market data on the cost reality of agents, whose pricing section was updated with it. Every addition passed the human gate before it reached this page.

The white paper became a knowledge graph. All six sides of the cube are published: the Harness, Architecture, Reliability and Security, Governance, Economics and Organization, seeded from the paper's full text with its 96 verified references carried over as sources and 98 claims traceable to them. Every major node descends through the level grammar: an anatomy figure, an operations figure with clickable regions, prose with anchoring figures, and receipts.

From this point on, the site is maintained by the living pipeline: an agent scans the territory, proposes changes on a branch, a compiler gate checks every proposal, and a human reviews what survives. Changes that pass the gate earn an entry on this page.